Xingyue Huang

3rd Year DPhil in Computer Science, University of Oxford. I like Graph neural networks, knowledge graphs, and graph representation learning in general.

[Ayhan Suleymanzade @ayhozade]

In conclusion, our results suggest that lossless superposition as local learning targets constitutes a sufficient condition for strong and efficient latent continuous reasoning.

📄Paper (PDF): https://t.co/sCLZ0E1WKw
🌐Project page: https://t.co/62NNKn2cXe
💻Code: https://t.co/b86SHGEzGf
⏳arXiv link is on hold, but coming hopefully soon!

8/🧵

[Ayhan Suleymanzade @ayhozade]

Can LLMs reason in superposition? We introduce MUX, a method that turns text CoT into latent continuous reasoning.

Instead of one-hot vectors as in CoT, the model now learns to predict weighted averages of several one-hot vectors, that we call multiplexed tokens. These multiplexed tokens can be designed to be lossless, so by predicting them one is essentially doing multi-token prediction (MTP) in superposition.

MUX is the best latent reasoning method across 32 math settings spanning 1-8B LLaMA base models, reducing CoT length by 3-6x. Furthermore, it is able to perform parallel search, harnessing a core strength of superposed reasoning.

In collaboration with @alperen_gozeten , @mmbronstein, @ismaililkanc, and @jw9730.

1/🧵

[Xingyue Huang @hxyscott]

Website: https://t.co/RNqmEAIdJ8

[Xingyue Huang @hxyscott]

Decisions for the Workshop on Graph Foundation Models: A New Era for Graph Machine Learning (GFM @ ICML 2026) have now been released on OpenReview.

This year, we received 78 submissions. Following a highly competitive review process, 62% of submissions were accepted.

[Xingyue Huang @hxyscott]

We are happy to announce that "Graph Foundation Models: A New Era for Graph Machine Learning" workshop has been accepted at #ICML2026!

Call for papers on topics: Graph Foundation Models, LLMs/TFMs+Graphs, theory of GFMs, etc.

📷 Submit by May 3rd! 📷 #GraphML #GFM

[Xingyue Huang @hxyscott]

A huge thank you to all authors, reviewers, and area chairs for helping make the workshop possible.

We are excited to see the growing momentum around graph foundation models and graph-centric AI systems.

We look forward to seeing everyone at ICML 2026 in Seoul!

[CAMEL-AI.org @CamelAIOrg]

🚨CAMEL-AI Live Talk by Xingyue Huang @hxyscott on RelAgent, an LLM-agent framework that acts as a data scientist for relational learning.

RelAgent autonomously searches via tool-calling over SQL feature programs and downstream models, guided by validation feedback and an evaluation workspace for task-specific diagnostics.

At inference time, there are no LLM calls, only executable SQL features + a classical model. RelAgent achieves competitive results against relational foundation models while staying fast, deterministic, and interpretable.
Paper: https://t.co/TyaX9fA8I6

⏲️ 15 May 2026 13:00 BST | 8:00 EDT
🔗 Register: https://t.co/DYpPLCvOk6

[Xingyue Huang @hxyscott]

7/🧵 RelAgent points to an alternative direction for relational learning:

Use LLMs not to replace relational models, but to help construct them.

Paper Link: https://t.co/TNfcPCnTa9
Github: https://t.co/tRXJFtMNn8

Built with @CamelAIOrg @duckdb @RelBench

[Xingyue Huang @hxyscott]

Can LLMs help relational learning while avoiding high inference cost?

Introducing RelAgent: an LLM agent that searches over SQL feature programs and classical ML models, then deploys the predictor without further LLM calls.

With @louistichelman @jw9730 @kolejnyyyy @ismaililkanc

[Xingyue Huang @hxyscott]

6/🧵 Empirically, RelAgent performs strongly across RelBenchV1, RelBenchV2, and 4DBInfer.

It is competitive with supervised tabular pipelines, relational deep learning models, relational foundation models, and LLM-as-feature-engineering baselines.

[Xingyue Huang @hxyscott]

Now that the NeurIPS deadline is over, consider submitting your work to the ICML 2026 Workshop on Graph Foundation Models (GFM)!

We would love to see more exciting work from the community!

🗓️ Submission Deadline: May 8, 2026 AoE

[Xingyue Huang @hxyscott]

📢 Deadline Extension Announcement — ICML 2026 GFM Workshop

To better accommodate the NeurIPS submission timeline, we have extended the submission deadline.

🗓️ New Deadline: May 8, 2026

We hope this extension provides additional flexibility for authors.

[Xingyue Huang @hxyscott]

📢 Deadline Extension Announcement — ICML 2026 GFM Workshop

To better accommodate the NeurIPS submission timeline, we have extended the submission deadline.

🗓️ New Deadline: May 8, 2026

We hope this extension provides additional flexibility for authors.

[Xingyue Huang @hxyscott]

We are happy to announce that "Graph Foundation Models: A New Era for Graph Machine Learning" workshop has been accepted at #ICML2026!

Call for papers on topics: Graph Foundation Models, LLMs/TFMs+Graphs, theory of GFMs, etc.

📷 Submit by May 3rd! 📷 #GraphML #GFM

[Michael Bronstein @mmbronstein]

Afternoon poster session

[Xingyue Huang @hxyscott]

If you’re around, come say hi and chat about graph foundation models, hypergraphs, and relational learning.

[Xingyue Huang @hxyscott]

Excited to be in Rio for ICLR 2026 🇧🇷

Tomorrow afternoon:
• HYPER: A Foundation Model for Inductive Link Prediction with Knowledge Hypergraphs
📍 Pavilion 3 — P3-#902

• Flock: A Knowledge Graph Foundation Model via Learning on Random Walks
📍 Pavilion 4 — P4-#5309

[Xingyue Huang @hxyscott]

We are also seeking sponsors, ACs, and reviewers! Please send us an email if you are interested!

[Xingyue Huang @hxyscott]

We are happy to announce that "Graph Foundation Models: A New Era for Graph Machine Learning" workshop has been accepted at #ICML2026!

Call for papers on topics: Graph Foundation Models, LLMs/TFMs+Graphs, theory of GFMs, etc.

📷 Submit by May 3rd! 📷 #GraphML #GFM

[Xingyue Huang @hxyscott]

Website: https://t.co/RNqmEAIdJ8
Openreview: https://t.co/uwvMhc0hmj